Design Of Dynamic Multi-Factor Quantization Strategy Based On Random Forest

Today,with the rapid development of big data technology,various data processing technologies and machine learning algorithms are applied to the financial industry,and quantitative investment emerges at the historic moment.In the background of the continuous promotion of financial market reform,quantitative investment is showing its great attraction and development space,and more and more investors in the domestic market are more inclined to choose the advantages of systematic and rigorous quantitative trading and investment methods.From the major fund product income ranking,quantitative long product income is indeed significantly higher than traditional value investment.A complete quantitative investment strategy should include stock selection,timing,position management and profit and loss control,among which the most important is to select quality stocks and determine the timing of stock trading.In order to determine the stock selection model more rigorously,this paper takes the multi-factor model as the theoretical support,builds two stock selection models based on this model,namely traditional scoring method and random forest,and compares and analyzes the stock selection backtest results of the two strategies.The empirical results show that,Compared with traditional stock selection method,random forest has more obvious advantages when dealing with a large number of factors due to its nonlinear learning ability.The strategy return of using the stochastic forest algorithm to build a portfolio is 26.26%,Higher than the scoring method of stock selection strategy yield 16.72%,Therefore,the stochastic Forest algorithm is preliminarily selected for multifactor stock selection.In order to further improve the profitability and stability of the model,the AMA adaptive timing model is superplaced on the basis of the random forest stock selection model,and the buying and selling points of the selected stock pool are determined according to the AMA moving average,that is,buying when the trend is up and selling when the trend is down.Compared with the backtest results,it was found that although the strategy return increased significantly to 32.3% after superposition of timing model,the overall return was 23.09%,and the stability of the strategy needed to be improved.Therefore,the "northbound funds" factor is further introduced to calculate the daily net inflow amount of "northbound funds",and then the 10-day average of the net inflow amount is calculated on a rolling basis.If the day’s net northbound inflows are less than its average for the past 10 days,it means the stop loss point has been reached,at which point you should sell all your holdings to control risk and buy the corresponding shares until the next trading signal.The backtest results show that the annualized return of the whole strategy is 33.16% and the maximum retracement is reduced to 11.54% after adding the "northbound money" factor to stop the loss,which confirms the effectiveness of this step and further proves the rationality of "northbound money" called "smart money".Therefore,investors can take it as a reliable investment reference signal to consider.